Discrete sounds better than integrated? a possible reason

[dbelousov]

The same with the "null tests". If I hear some difference that is not measurable with common metrics that are available to me (FR, THD, etc.) I just make a null test with some real and complicated actual music.

For mics, it is not easy to implement. So I use a different strategy: a reference mic. I put a very decent microphone as the "reference point" and the mic I'm working on is always compared to it with every adjustment.

 

[abbey road d enfer]

Yes.
Well I'm pretty sure Sinatra/Elvis would sound good even through an SM57

that's exactly it. Too much fetichism about gear.

 

[JohnRoberts]

I think this should be required reading for anyone designing with op amps like the TL072:

https://www.ti.com/lit/an/sloa020a/sloa020a.pdf

Cheers

Ian

OK as promised here's today's TMI about TL07x and NF.

Last century I killed way too many brain cells trying to improve phono preamps that were already way better than vinyl. To connect the TL07x dot and NF dot, a preamp kit I designed in the early 1980s used a Cohen topology flat gain front end (JFET for MM and low noise bipolar for MC), followed by a novel RIAA EQ topology.

Without getting too esoteric, but warning what follows is pretty esoteric, as I investigated in my 1980 article about console design the typical op amp input error voltage (effectively 1/loop gain) is phase shifted 90' by the dominant pole stability compensation above that pole frequency.  This means that the input error voltage is phase shifted 90' wrt to input and output signals. This is generally innocuous as long as loop gain margin is large so error voltage is relatively small causing only a tiny net phase shift in practice. Operating such op amps at too high noise gains (like virtual earth sum bus amp with too many stems) can result in measurable phase shift and more. :-( 

Getting back to my RIAA phono preamp, the RIAA EQ involve a pole at 3,180uSec, a zero at 318 uSec, and another pole at 75 uSec. Looking at the bode response, open loop gain plot of TL07x (figure 9 in Ian's link), we observe that the dominant pole compensation in the TL07x begins lower than the 3,180 uSec RIAA pole. This means that the TL07x configured as a LPF integration stage would exhibit input error voltage that would be effectively in phase with the closed loop transfer function. I warned you that this was more than a little esoteric.

Here is a link to that preamp kit article with schematic. http://www.johnhroberts.com/p-10_art.pdf

I make no other claims about this other than it is way too much information about NF and coincidentally about TL07x that I have used truckloads of over the decades...

JR 

 

[Winston OBoogie]

Thing is, discrete circuitry needn't sound like anything. 
Again it's down to how things misbehave.

If you're wanting transparent, you'd have to be pretty good with discretes and know your stuff to better something like the 5532.  It is possible but, there's a lot to consider. 

Where an I.C. fails in certain applications is:

1/ there aren't any that I know of with big beefy BJT or J-Fet  diff. amps that can compete, noise wise, with the sub 1 nv/rt. Hz of the old Rohm & Toshiba's. So you're looking at putting another I.C. of matched low noise BJT's before it for mic amp applications etc.

2/ Put a low F.B load on an I.C. such as the 5532 (again, for noise reasons), and it'll be operating in class B.  Well designed class B but, class B nevertheless. 
I'd rather a properly biased class B than AB,.  But better still is class A.  So we put a class A buffer on the output, usually a discrete, but there are I.C. 's for that too.  Current source the 5532 output to about 2mA class A and run the thing into your high Z buffer of choice.

It's pretty easy to build a discrete that misbehaves in some way that we perceive as nice.  But damn hard to build one that's bullet proof and "blameless". 

Another thing to consider is that discrete allows the circuit to do exactly what we want, and where we want.  In some circumstances, a standard "building block" 5532 or TL072 is too much or not enough.
But again, you'd better be damn good with discrete to better what you can do with the above standards in terms of blamelessness and transparency. 

 

[iampoor1]

It's pretty easy to build a discrete that misbehaves in some way that we perceive as nice.  But damn hard to build one that's bullet proof and "blameless". 

I have often wondered if a big part of the general opinion that discrete = sound better is that dont discrete circuits tend to clip asymmetrically with respect to the Positive and negative rails?

 

[Winston OBoogie]

I have often wondered if a big part of the general opinion that discrete = sound better is that dont discrete circuits tend to clip asymmetrically with respect to the Positive and negative rails?

Some do.  Some don't.  All kinds of misbehaviors -  slew rate limiting, VAS/TIS clipping before output stage, simple compensation schemes that limit fb at H.F.,  ... anything really.

You could design a discrete with staggered rails, off the charts headroom on all stages, whatever you desire.
I'm just trying to make the point that discrete needn't sound like anything.  There is no one discrete sound in my opinion.

Keep in mind that a lot of the circuits folks clone/copy on here were designed when transistors were expensive and still fairly new.  So you see a lot of 3 transistor gain stages.  This was really down to economy, of money and possibly knowledge?  Also borne out of working with tubes so you see similar topologies being adopted with early Germanium & BJT

For the record, I'm not at all knocking these early efforts. 
It's all good  8)

 

[ruffrecords]

Hello

The nevegate  ;D
Faked THD emission measurement for aproval  8)
I guess the one programming the software that recognise the test bench situation and minimise compromising data will have trouble wih regulation

Best
Zam

LOL. Storm in a tea cup. The object was to verify correct operation of the Neve parts of the mixer not the distortion of the VU meter.

Cheers

Ian

 

[abbey road d enfer]

The object was to verify correct operation of the Neve parts of the mixer not the distortion of the VU meter.

But were customers informed of this degradation? I don't remember having seen such caveat when I was in charge of the infamous Barclay desk.
Why didn't Neve provide meter buffers?

 

[ruffrecords]

But were customers informed of this degradation? I don't remember having seen such caveat when I was in charge of the infamous Barclay desk.
Why didn't Neve provide meter buffers?

They did provide the option of regular VU meters, buffered ones or PPMs (Pete Townshend had PPMs on his Neve desk). I do not know why but most people chose unbuffered VUs. Unbuffered VUs were common everywhere back in the 70s when I was there.

Cheers

Ian

 

[abbey road d enfer]

They did provide the option of regular VU meters, buffered ones or PPMs (Pete Townshend had PPMs on his Neve desk). I do not know why but most people chose unbuffered VUs. Unbuffered VUs were common everywhere back in the 70s when I was there.

Cheers

Ian

It would be interesting to know if this choice was purely economic, or distortion was considered negligible (or even desirable), or if it was pure ignorance.

 

[L´Andratté]

Thanks Ruffrecords and all for discussing this!

This is a mic pre I built with some old Sowter 1:5 transformers (it is a design by Ricardo from a somewhat unnerving GDIY thread, but it is almost exactly to be found in AoE 2nd edition. IIRC GDIY member joechris made me aware of it).
It´s not a NEVE, but very exemplary in "sounding discreeete" :, soft onset of distortion, little harshness, etc. No output buffer for extra loading, it´s a classy low noise distortion pedal...
Of course the transformers are in there too so as ever, grain of salt, etc.

Maybe someone likes to make a point about it.
Maybe not. It´s all I could possibly contribute--

https://groupdiy.com/index.php?topic=64069.msg811214#msg811214

 

[Winston OBoogie]

That circuit of Ricardo's looks good for only using 2  transistors. 
He didn't even bother with the usual emitter follower before Q2 
I like it , I bet it sounds nice on a lot of things. 

Edit:  On reflection, this is typical for early '70's stuff like the Helios pre etc., except, I bet it's a lot quieter.

Very small current (94uA) on the input Q1 which would be about perfect for a 1:5 mic input transformer.
I'm assuming the 3K8 resistor shown as R12 is not really there on the actual circuit but is shown to reflect what the 1:5 transformer will present. 

 

[L´Andratté]

Yes, it´s looking into the secondary

 

[Winston OBoogie]

Yes, it´s looking into the secondary

Yep thanks.

It might have helped had I read the thread you linked to  ???
I have now 

Again, nice.  Thanks for posting.

 

 

[abbey road d enfer]

On reflection, this is typical for early '70's stuff like the Helios pre etc., except, I bet it's a lot quieter.

Why would that be? I reckon a BC109C at the input would not be very different in terms of noise.

 

[Winston OBoogie]

Why would that be? I reckon a BC109C at the input would not be very different in terms of noise.

It isn't the device, it's the fact it's optimized for lowest noise with the source impedance.

You mention the BC109 which is what Neve used early on.
They (Neve) couldn't optimize for lowest noise because the impedance seen by the first BC109 varies depending on attenuator setting. 
This because they chose to not vary the amp gain too much and opted to go for a divider network on the IP transformer secondary.  So the current in the input transistor is at a compromise setting.

On Helios, they used a honking 1:10 input transformer and, with that reflected secondary impedance, we're really into Toshiba extinct J-Fet territory for lowest noise.


Edit:  Impedance seen by input transistor in Neve 10** series modules varies from, circa sub 60 ohms in line input mode, to about 1K4 with attenuator halfway down in mic mode.  This from memory so, subject to checking of maths

 

[ruffrecords]

It would be interesting to know if this choice was purely economic, or distortion was considered negligible (or even desirable), or if it was pure ignorance.

That is a good question but probably hard to answer. I know back in the 70s it was common to connect VU meters directly across outputs. It happened in mixers, in tape machines and external FX boxes. In the course of tracking a mastering in those days, the signal probably experienced several VU meters before it reached the recording medium. One the other hand, in those days most professional gear was designed to be able to drive a 600 ohm load so they had pretty low output impedances - which means they might be less affected by the non linear load of a VU than you might expect.

And on the third hand, I do not think many end users were aware of the distortion that VU meters added.

Cheers

Ian

 

[abbey road d enfer]

It isn't the device, it's the fact it's optimized for lowest noise with the source impedance.

You mention the BC109 which is what Neve used early on.
They (Neve) couldn't optimize for lowest noise because the impedance seen by the first BC109 varies depending on attenuator setting. 
This because they chose to not vary the amp gain too much and opted to go for a divider network on the IP transformer secondary.  So the current in the input transistor is at a compromise setting.

On Helios, they used a honking 1:10 input transformer and, with that reflected secondary impedance, we're really into J-Fet territory for lowest noise.


Edit:  Impedance seen by input transistor in Neve 10** series modules varies from, circa sub 60 ohms in line input mode, to about 1K4 with attenuator halfway down in mic mode.  This from memory so, subject to checking of maths

I understand all that, but your remark, IIUC, was addressed only to the posted schemo.
Regarding the Helios and its 1:10 xfmr, the preamps in my mixers used a similar one (Melodium, if you remember them) with a BC109C operating at 50uA.
I measured -128dBu input noise with a 200r dummy.
About the Neve, the operating point must be chosen for the highest gain, since S/N ratio can only improve, unless someone does crazy things with attenuators.
I have designed a preamp where the input device (BC560) runs at 200uA and normally sees a 2k secondary. I dedicate it to drums and such brutal sources. It is not the quietest in the world but nobody complained so far.
The input pad is based on an alternate primary (20k nominal), so the transistor base sees about 20 ohms+ some DCR, so talk about mismatch! Anyway, the resulting S/N ratio is perfectly manageable and inobtrusive.
And yes, the output stage is class A. 

 

[Winston OBoogie]

That is a good question but probably hard to answer. I know back in the 70s it was common to connect VU meters directly across outputs. It happened in mixers, in tape machines and external FX boxes. In the course of tracking a mastering in those days, the signal probably experienced several VU meters before it reached the recording medium. One the other hand, in those days most professional gear was designed to be able to drive a 600 ohm load so they had pretty low output impedances - which means they might be less affected by the non linear load of a VU than you might expect.

And on the third hand, I do not think many end users were aware of the distortion that VU meters added.

Cheers

Ian

I'm guilty of advising the testers at UA to test distortion on the LA-2A with the meter in gain reduction mode. 
They'd been doing it on the AP after checking line up levels etc.  and the meter was usually left in output monitoring mode.
I took a unit over to my FFT setup and showed them the effect of the meter on the distortion spectra.
It's quite clearly there even with a 600 om load.


Now in my defense, most folks will be using an LA-2A monitoring in gain reduction but, for my sins, flame away!

 

[abbey road d enfer]

And on the third hand, I do not think many end users were aware of the distortion that VU meters added.

Cheers

Ian

And it's a pity. A long time ago, I demonstrated that to a colleague. I showed him the distortion residues on the scope and was horrified.
I understand the motivations for doing without a meter buffer, cost, complexity, more things that can go wrong, lack of confidence in a measurement that is not done on the actual signal, but the deterioration of performance is not negligible, even in times when tape was the dominant distortion factor.